AI Insight
Researchers developed HERO, a new AI system that improves automated radiology report generation by using a hierarchical optimization approach that separates reasoning, diagnosis, and evidence grounding into distinct optimization levels. The system addresses a key limitation in existing AI models where uniform credit assignment across generated text leads to clinical hallucinations and misalignment between evidence and diagnoses. Testing on two major medical imaging datasets (MIMIC-CXR and IU-Xray) showed HERO outperformed existing methods in diagnostic accuracy while producing reports with better evidence grounding and internal consistency.
Why it matters
This approach could help radiologists by generating more reliable and clinically accurate automated reports, potentially reducing workload and improving diagnostic consistency. The method's focus on reducing hallucinations and maintaining evidence-diagnosis alignment addresses critical safety concerns in clinical AI applications.
Understand the Science
⚠️ Preprint – Noch nicht peer-reviewed
Dieser Artikel wurde noch nicht von unabhängigen Experten begutachtet. Die Ergebnisse sind vorläufig und sollten mit Vorsicht interpretiert werden.
Abstract: Multimodal Large Language Models (MLLMs) have substantially advanced Radiology Report Generation (RRG), yet aligning them through reinforcement learning (RL) remains challenging due to heterogeneous medical supervision. Vanilla Group Relative Policy Optimization (GRPO) assigns uniform credit across the entire generation, leading to segment interference, token dilution, and evidence–diagnosis decoupling, which exacerbates clinical hallucinations. We propose HERO (Hierarchical Evidential Reasoning Optimization), a factorized policy optimization framework that aligns heterogeneous supervision with three optimization granularities. HERO separately optimizes reasoning, diagnosis, and evidence grounding through complementary segment-, token-, and completion-level optimization with a heterogeneous reward formulation covering diagnostic accuracy, reasoning quality, and think–answer consistency. Experiments on MIMIC-CXR and IU-Xray show that HERO outperforms strong supervised and reinforcement learning baselines, achieving state-of-the-art clinical efficacy while producing more evidence-grounded and think–answer-consistent reports, thereby substantially mitigating clinical hallucinations.